Pulmonary Fissure Detection In Three-dimensional High Resolution CT

Pulmonary disease has an increased incidence,serious harm to human health,common lung disease including: pneumothorax,emphysema,pneumonia and lung cancer.Among them,lung cancer is the most common primary malignant tumor,which is one of the highest mortality of the disease,early discovery of lung lesions,the patient can improve the survival rate.High resolution CT of chest has been widely used to detect and diagnose pulmonary disease in clinical medicine.Pulmonary fissure divide human lungs into five regions called lobes.The accurate identification of the fissure is considerably important for the location of pulmonary lesions and the precise analysis of lung function,which can assist the development of preoperative planning and postoperative evaluation in clinical practice.However,automated detection of pulmonary fissures in CT image is not an easy task,as fissures are soft tissues which appear as thin,weak and varying structures,often undulating planar surfaces with blurred boundaries.This is further complicated by presence of surrounding vessels and other structures.Therefore,this paper first preprocess lung image,avoiding the interference from irrelevant tissues;then enhancing fissures to expand the difference between pulmonary fissures and other structures,inhibiting characteristics are not interested;Finally,a shortest-path algorithm is used to segment fissure completely.The main work includes:(1)The fissure profile typically appear as a bright thin curviline across the section,by using this feature,we propose a derivative of stick(DoS)filter for fissure enhancement based on preprocessing image.The adjacent background on both sides of the fissure takes a low-intensity narrow ribbon shape under CT scanning.Taking this special intensity distribution into account,we propose to probe the fissure profile and its close neighborhood using a template composed of three parallel sticks.This is exactly the same principle which our human observers use to infer the presence of fissure especially fo r weak objects and inhomogeneous intensity,which makes it different from the conventional isotropic filters and can help to get more robust response even under serious noise and axial discontinuity.(2)We developed a semi-automatic segmentation method using Dijkstra shortest path algorithm based on the enhanced image.First,The local cost function is a weighted sum of the component cost functionals for each of the following features.Features include gradient magnitude,gradient direction and the laplacian zero-crossing,making the fissure path to be the shortest;then manually select the pixels belonging to the fissure as the source point in sagittal plane,using Dijkstra shortest path algorithm to compute the shortest path to the other pixels.The exploratory Dijkstra algorithm performed in three perpendicular cross-sectional iteratively,reducing the user 's input possibly.We improve the traditional single-source,single-destination Dijkstra algorithm to the single source,multi-destination,can effectively detect fissures and keep track of its thickness and integrity;Finally,in order to further improve the segmentation results,using morphological post-processing operation to remove the interference of extraneous tissue,getting a smooth segmentation results of lung fissure.The performance of our methods has been verified in experiments using chest CT scans of 23 Chronic Obstructive Pulmonary Disease patients.We define two evaluation indicators: the accuracy ACC and the false discovery rate FDR.By contrast with Ground Truth shows that,the proposed algorithm has better segmentation performance,fully meet the needs of clinical applications.